Search

The exponential growth of user-generated content has increased the need for efficient video summarization schemes. However, most approaches underestimate the power of aural features, while they are designed to work mainly on commercial/professional videos. In this work, we present an approach that uses both aural and visual features in order to create video summaries from user-generated videos. Our approach produces dynamic video summaries, that is, comprising the most “important” parts of the original video, which are arranged so as to preserve their temporal order. We use supervised knowledge from both the aforementioned modalities and train a binary classifier, which learns to recognize the important parts of videos. Moreover, we present a novel user-generated dataset which contains videos from several categories. Every 1 s part of each video from our dataset has been annotated by more than three annotators as being important or not. We evaluate our approach using several classification strategies based on audio, video and fused features. Our experimental results illustrate the potential of our approach.

Inertial measurement units (IMUs) enable orientation, velocity, and position estimation in several application domains ranging from robotics and autonomous vehicles to human motion capture and rehabilitation engineering. Errors in orientation estimation greatly affect any of those motion parameters. The present work explains the main challenges in inertial orientation estimation (IOE) and presents an extensive benchmark dataset that includes 3D inertial and magnetic data with synchronized optical marker-based ground truth measurements, the Berlin Robust Orientation Estimation Assessment Dataset (BROAD). The BROAD dataset consists of 39 trials that are conducted at different speeds and include various types of movement. Thereof, 23 trials are performed in an undisturbed indoor environment, and 16 trials are recorded with deliberate magnetometer and accelerometer disturbances. We furthermore propose error metrics that allow for IOE accuracy evaluation while separating the heading and inclination portions of the error and introduce well-defined benchmark metrics. Based on the proposed benchmark, we perform an exemplary case study on two widely used openly available IOE algorithms. Due to the broad range of motion and disturbance scenarios, the proposed benchmark is expected to provide valuable insight and useful tools for the assessment, selection, and further development of inertial sensor fusion methods and IMU-based application systems.

Spectral unmixing represents both an application per se and a pre-processing step for several applications involving data acquired by imaging spectrometers. However, there is still a lack of publicly available reference data sets suitable for the validation and comparison of different spectral unmixing methods. In this paper, we introduce the DLR HyperSpectral Unmixing (DLR HySU) benchmark dataset, acquired over German Aerospace Center (DLR) premises in Oberpfaffenhofen. The dataset includes airborne hyperspectral and RGB imagery of targets of different materials and sizes, complemented by simultaneous ground-based reflectance measurements. The DLR HySU benchmark allows a separate assessment of all spectral unmixing main steps: dimensionality estimation, endmember extraction (with and without pure pixel assumption), and abundance estimation. Results obtained with traditional algorithms for each of these steps are reported. To the best of our knowledge, this is the first time that real imaging spectrometer data with accurately measured targets are made available for hyperspectral unmixing experiments. The DLR HySU benchmark dataset is openly available online and the community is welcome to use it for spectral unmixing and other applications.

Substantial efforts have been devoted more recently to presenting various methods for object detection in optical remote sensing images. However, the current survey of datasets and deep learning based methods for object detection in optical remote sensing images is not adequate. Moreover, most of the existing datasets have some shortcomings, for example, the numbers of images and object categories are small scale, and the image diversity and variations are insufficient. These limitations greatly affect the development of deep learning based object detection methods. In the paper, we provide a comprehensive review of the recent deep learning based object detection progress in both the computer vision and earth observation communities. Then, we propose a large-scale, publicly available benchmark for object DetectIon in Optical Remote sensing images, which we name as DIOR. The dataset contains 23,463 images and 192,472 instances, covering 20 object classes. The proposed DIOR dataset (1) is large-scale on the object categories, on the object instance number, and on the total image number; (2) has a large range of object size variations, not only in terms of spatial resolutions, but also in the aspect of inter- and intra-class size variability across objects; (3) holds big variations as the images are obtained with different imaging conditions, weathers, seasons, and image quality; and (4) has high inter-class similarity and intra-class diversity. The proposed benchmark can help the researchers to develop and validate their data-driven methods. Finally, we evaluate several state-of-the-art approaches on our DIOR dataset to establish a baseline for future research. [ABSTRACT FROM AUTHOR]

Image saliency detection is an active research topic in the community of computer vision and multimedia. Fusing complementary RGB and thermal infrared data has been proven to be effective for image saliency detection. In this paper, we propose an effective approach for RGB-T image saliency detection. Our approach relies on a novel collaborative graph learning algorithm. In particular, we take superpixels as graph nodes, and collaboratively use hierarchical deep features to jointly learn graph affinity and node saliency in a unified optimization framework. Moreover, we contribute a more challenging dataset for the purpose of RGB-T image saliency detection, which contains 1000 spatially aligned RGB-T image pairs and their ground truth annotations. Extensive experiments on the public dataset and the newly created dataset suggest that the proposed approach performs favorably against the state-of-the-art RGB-T saliency detection methods. [ABSTRACT FROM AUTHOR]

This paper presents a software library as a research and educational tool for Multi-Skill Resource-Constrained Scheduling Problem. The following useful tools have been implemented in Java: instance Generator, solution validator, solution visualizer and example solvers: Greedy algorithm and Genetic Algorithm. All tools are supported by iMOPSE dataset which consists of 36 instances and additional 'small' 6 instances for educational purpose. In the paper, three test studies are described: (1) educational use of 6 'small' instances, (2) optimization of cost or duration of a schedule, and (3) simple bicritieria optimization of cost/duration of a final schedule. All described tools/examples are freely published on iMOPSE homepage. [ABSTRACT FROM AUTHOR]

Deep learning-based visual perception of an unmanned surface vehicle (USV) essentially benefits from informative images. It is unfortunately short of available datasets of marine vessel targets for visual perception system of USVs. In this paper, a marine vessel detection dataset, termed MVDD13, is exclusively established by deploying 35,474 images which are accurately annotated to be 13-category vessels. The proposed MVDD13 is equipped with benchmark features by sufficiently taking into account the reality of category proportion, image diversity, sample independence, and background confusion, etc. , thereby facilitating rather deep information for training and testing a robust detector. To evaluate benchmark quality, the state-of-the-art deep learning detectors conducting on the MVDD13 show that individual advantages can be consistently reflected. Furthermore, by cross-dataset training/testing, intensive comparisons on general COCO and specific SeaShips datasets reveal that the MVDD13 can significantly fertilize deep learning-based recognition performance in terms of mAP@.5, mAP@.75 and mAP@[.5:.95]. All images and annotations can be downloaded via https://github.com/yyuanwang1010/MVDD13. [ABSTRACT FROM AUTHOR]

License plate recognition plays a critical role in many practical applications, but license plates of large vehicles are difficult to be recognized due to the factors of low resolution, contamination, low illumination, and occlusion, to name a few. To overcome the above challenges, the transportation management department generally introduces the enlarged license plate behind the rear of a vehicle. However, enlarged license plates have high diversity as they are non-standard in position, size, and style. Furthermore, the background regions contain a variety of noisy information which greatly disturbs the recognition of license plate characters. In this work, we address the enlarged license plate recognition problem and contribute a dataset containing 9342 images, which cover most of the challenges of real scenes. However, the created data are still insufficient to train deep methods of enlarged license plate recognition, and building large-scale training data is very time-consuming and high labor cost. To handle this problem, we propose a novel data generation framework based on the Disentangled Generation Network (DGNet), which disentangles the generation of enlarged license plate data into the text generation and background generation in an end-to-end manner to effectively ensure the generation diversity and integrity, for robust enlarged license plate recognition. Extensive experiments on the created dataset are conducted, and we demonstrate the effectiveness of the proposed approach in three representative text recognition frameworks. • We contribute a dataset containing 9342 images for enlarged license plate recognition (ELPR). • We propose a novel end-to-end text and background generation framework for ELPR. • We design several schemes to ensure the quality of generated enlarged license plates. • We evaluate the effectiveness of our method using three representative frameworks. [ABSTRACT FROM AUTHOR]

To tackle the challenge of vehicle re-identification (Re-ID) in complex lighting environments and diverse scenes, multi-spectral sources like visible and infrared information are taken into consideration due to their excellent complementary advantages. However, multi-spectral vehicle Re-ID suffers cross-modality discrepancy caused by heterogeneous properties of different modalities as well as a big challenge of the diverse appearance with different views in each identity. Meanwhile, diverse environmental interference leads to heavy sample distributional discrepancy in each modality. In this work, we propose a novel cross-directional consistency network (CCNet) to simultaneously overcome the discrepancies from both modality and sample aspects. In particular, we design a new cross-directional center loss (L c d c) to pull the modality centers of each identity close to mitigate cross-modality discrepancy, while the sample centers of each identity close to alleviate the sample discrepancy. Such a strategy can generate discriminative multi-spectral feature representations for vehicle Re-ID. In addition, we design an adaptive layer normalization unit (ALNU) to dynamically adjust individual feature distribution to handle distributional discrepancy of intra-modality features for robust learning. To provide a comprehensive evaluation platform, we create a high-quality RGB-NIR-TIR multi-spectral vehicle Re-ID benchmark (MSVR310), including 310 different vehicles from a broad range of viewpoints, time spans and environmental complexities. Comprehensive experiments on both created and public datasets demonstrate the effectiveness of the proposed approach comparing to the state-of-the-art methods. The dataset and code will be released for free academic usage at https://github.com/superlollipop123/Cross-directional-Center-Network-and-MSVR310. • A cross-directional consistency network for multi-spectral vehicle re-identification. • A cross-directional center loss to simultaneously pull modality and sample centers. • An adaptive layer normalization to adjust feature distribution in each modality. • A high-quality multi-spectral vehicle re-identification benchmark dataset MSVR310. [ABSTRACT FROM AUTHOR]

Deep learning has achieved great success on robotic vision tasks. However, when compared with other vision-based tasks, it is difficult to collect a representative and sufficiently large training set for six-dimensional (6D) object pose estimation, due to the inherent difficulty of data collection. In this paper, we propose the RobotP dataset consisting of commonly used objects for benchmarking in 6D object pose estimation. To create the dataset, we apply a 3D reconstruction pipeline to produce high-quality depth images, ground truth poses, and 3D models for well-selected objects. Subsequently, based on the generated data, we produce object segmentation masks and two-dimensional (2D) bounding boxes automatically. To further enrich the data, we synthesize a large number of photo-realistic color-and-depth image pairs with ground truth 6D poses. Our dataset is freely distributed to research groups by the Shape Retrieval Challenge benchmark on 6D pose estimation. Based on our benchmark, different learning-based approaches are trained and tested by the unified dataset. The evaluation results indicate that there is considerable room for improvement in 6D object pose estimation, particularly for objects with dark colors, and photo-realistic images are helpful in increasing the performance of pose estimation algorithms.

Highly accurate and precise heave decay tests on a sphere with a diameter of 300 mm were completed in a meticulously designed test setup in the wave basin in the Ocean and Coastal Engineering Laboratory at Aalborg University, Denmark. The tests were dedicated to providing a rigorous benchmark dataset for numerical model validation. The sphere was ballasted to half submergence, thereby floating with the waterline at the equator when at rest in calm water. Heave decay tests were conducted, wherein the sphere was held stationary and dropped from three drop heights: a small drop height, which can be considered a linear case, a moderately nonlinear case, and a highly nonlinear case with a drop height from a position where the whole sphere was initially above the water. The precision of the heave decay time series was calculated from random and systematic standard uncertainties. At a 95% confidence level, uncertainties were found to be very low—on average only about 0.3% of the respective drop heights. Physical parameters of the test setup and associated uncertainties were quantified. A test case was formulated that closely represents the physical tests, enabling the reader to do his/her own numerical tests. The paper includes a comparison of the physical test results to the results from several independent numerical models based on linear potential flow, fully nonlinear potential flow, and the Reynolds-averaged Navier–Stokes (RANS) equations. A high correlation between physical and numerical test results is shown. The physical test results are very suitable for numerical model validation and are public as a benchmark dataset.

The semantic labeling of the urban area is an essential but challenging task for a wide variety of applications such as mapping, navigation, and monitoring. The rapid advance in Light Detection and Ranging (LiDAR) systems provides this task with a possible solution using 3D point clouds, which are accessible, affordable, accurate, and applicable. Among all types of platforms, the airborne platform with LiDAR can serve as an efficient and effective tool for large-scale 3D mapping in the urban area. Against this background, a large number of algorithms and methods have been developed to fully explore the potential of 3D point clouds. However, the creation of publicly accessible large-scale annotated datasets, which are critical for assessing the performance of the developed algorithms and methods, is still at an early age. In this work, we present a large-scale aerial LiDAR point cloud dataset acquired in a highly-dense and complex urban area for the evaluation of semantic labeling methods. This dataset covers an urban area with highly-dense buildings of approximately 1 km2 and includes more than three million points with five classes of objects labeled. Moreover, experiments are carried out with the results from several baseline methods, demonstrating the feasibility and capability of the dataset serving as a benchmark for assessing semantic labeling methods.